The sudden resurgence of AI in the early 2010s, riding on the coattails of newly developed and groundbreaking machine learning and deep learning algorithms, was accompanied by the emergence of seemingly endless “future of work” conversations and, eventually, heated debates .
By the end of the decade, generating target listings by job and/or sector had become a popular trend, with all trying to predict which percentage of total jobs could be automated. Potential productivity gains and levels of (un)employment were also part and parcel of the estimates. The unexpected success of ChatGPT and other large language models (LLMs) added more fuel to the fire. The knowledge-generation capabilities of the new LLMs have greatly expanded the range of jobs exposed, adding cognitive tasks and jobs previously spared by older technological innovations.
There are, however, two sides to the issue. One is how technologies such as AI impact jobs and employment, briefly summarized above. The other is how these technologies use existing labor to be produced. Given its complexity, AI demands a full dosage of highly skilled labor capable of mastering the new AI algorithms and platforms while keeping the innovation wave alive. Talent search thus became a priority, with Big Tech companies typically leading the charge, offering top salaries and even poaching workers from competitors.
Talent nurturing and development have also become a cornerstone of national AI strategies, usually set as a central pillar in most of them. The paradox here is that new AI technologies, now under the steering of world-class talent, create the conditions to eliminate the need to deploy part of such talent. Their new human-like capabilities enable them to successfully undertake many of the tasks at hand. Computer science and coding are perhaps the best examples in this regard.
But that is not the end of the story. Low-skill labor is also a crucial input into the overall production of digital technologies, in general, and AI in particular. While work environments in distribution centers, where low-skill laborers are closely monitored every second, resemble Bentham’s beloved Panopticon, now running on AI, data cleaning, data labeling, and microwork performed by low-cost labor do essential work. This is work that AI cannot undertake independently. .
Ghost work has been used by researchers to describe such labor, most of which is remunerated at the lowest levels possible—or goes unpaid in some cases. Unlike labor in previous technological disruptions, AI has at its disposal a seemingly unlimited supply of low-skilled laborers who are spread around the world. It thus has no nationality, unlike the owners and managers of the global digital factories. In any case, low-skilled labor is commonly not part of national AI strategies and remains largely invisible to most. It has also received relatively less attention in academic research. In this post, I will focus on the potential impact of AI on jobs and employment.
Innovation, a word with negative connotations until the end of the 18th Century, plays a central role in societies where capitalism prevails. Surely, firms of all sizes strive to outdo the competition and be ahead of the game by deploying the latest technologies. This allows them to lower costs, gain market share, and eliminate the weakest capitals in the production chain. Economists usually refer to such a process as technical change, and, if you really want to know, there is yet another huge debate about whether such a change is endogenous or exogenous—albeit not relevant to my purposes here. At any rate, technological innovation can also be used to describe such a process. Succinctly, technical change is an indelible imprint of such societies that those running profit-driven businesses cannot afford to neglect. They will inexorably melt away otherwise.
At the macro level, such an imprint has historically manifested through the various industrial revolutions. The first one ran on the coattails of the steam engine and its multiple refinements since the mid-17th Century. Today, we frequently hear about the impending or ongoing Fourth Industrial Revolution, driven by the rapid development of new Information and Communication Technologies (ICTs). And now AI has surged as its indisputable leader, threatening to conquer all the interstices of society.
Technical change usually entails replacing human labor with more sophisticated and advanced machinery and technologies. Automation is the result, while increases in labor productivity (output per worker) are often expected. However, it can also lead to systemic labor displacement, depending on the precise character of the technologies being deployed in production processes. It also typically entails dramatic changes in job descriptions, requiring updated workers’ skills to handle the new technology appropriately, either lower skills or more sophisticated ones, for managers and engineers. Technological unemployment and labor force deskilling are two of its most common symptoms.
All industrial revolutions have piggybacked on a set of technologies with exceptional characteristics. Indeed, the steam engine, electricity, and ICTs have been their prime movers, thus accelerating technical change. Such technologies have been subsumed under the rubric of General-Purpose Technologies (GPTs, not to be confused with Generative Pre-trained Transformers), which have three core and interrelated traits :
- Dynamism → Rapid development, paced diffusion
- Pervasiveness → Impacts most sectors, especially core ones
- Complementary innovation → Triggers innovation all around, propelling all above (recursiveness)
While GPTs drive systemic technical change and innovation, not all innovation stems from GPTs. Sectoral or thematic-focused innovations occur quite frequently but have limited scope. More often than not, they remain trapped within a few production domains. They thus lack GPT traits 2 and 3 mentioned above.
Undoubtedly, GPTs foster automation in core productive sectors. They also boost overall productivity, giving first movers (firms, corporations and/or countries) a clear competitive advantage that promptly translates into increased profits, revenues and geopolitical advantages. Historically, GPT development has been accelerated, but its diffusion throughout society has been relatively slow. That is undoubtedly the case for key sectors of national economies, especially in advanced economies, where such a process commonly happens first.
Whether Generative AI (GenAI) will have the same labor and employment impacts as older GPTs, or if it introduces new elements into the puzzle, remains to be determined. GenAI and Agentic AI (AgenAI) could indeed have skill-equalizing effects. On the other hand, cognitive job-displacement is a distinct possibility.
Raul
References
